Type 2 fuzzy system models with type 1 inference

Fuzzy System Modeling (FSM) is one of the most prominent tools to identify the behavior of highly nonlinear systems with uncertainty. The main goal of this thesis is to propose an objective methodology that can identify robust type 2 Takagi-Sugeno and Mizumoto fuzzy system models with high prediction power.; In this thesis, Fuzzy C-Means (FCM) clustering algorithm is used to identify the natural behavior of the system. The proposed type 2 fuzzy system model structure is a collection of embedded type 1 fuzzy system models identified for different values of level of fuzziness. The proposed type 2 FSM method have four major components: (1) Data preprocessing, (2) Structure identification, (3) Parameter tuning, and (4) Model validation.; Two new input selection algorithms are proposed in order to identify the significant input variables in data preprocessing step. The first stage of one of the proposed input selection algorithms uses a new functional-dependency driven feature filter in order to identify a set of candidate optimal input variable combinations.; As mentioned above, the FCM clustering algorithm will be used in order to identify the inference parameters of the proposed fuzzy system model structures. The proposed distance measure is used in order to remedy the problems of the FCM clustering algorithm due to Euclidean and Mahalanobis distance measures. In order to further improve the proposed distance measure and to further tune the system model, a weight term is added.; The proposed inference mechanisms and test data are used in order to validate the identified system models. In order to be able to deduce a model output by using the proposed type 2 fuzzy system model structures, a type-reduction step is added.; The proposed algorithms are applied on five well-known benchmark data sets and a real-life network traffic data set. In order to measure the effectiveness of the proposed algorithm, same benchmark systems are modeled by using other system modeling tools. Root mean square prediction error is used as the performance index. The overall results demonstrated that the proposed system modeling techniques can effectively identify the behavior of highly nonlinear systems.